Atmospheric effects in the remote sensing of phytoplankton pigments

Abstract

We investigate the accuracy with which relevant atmospheric parameters must be estimated to derive phytoplankton pigment concentrations (chlorophyll a plus phaeophytin a ) of a given accuracy from measurements of the ocean's apparent spectral radiance at satellite altitudes. The analysis is limited to an instrument having the characteristics of the Coastal Zone Color Scanner scheduled to orbit the Earth on NIMBUS-G. A phytoplankton pigment algorithm is developed which relates the pigment concentration (C) to the three ratios of upwelling radiance just beneath the sea surface which can be formed from the wavelengths (λ) 440, 520 and 550 nm. The pigment algorithm explains from 94 to 98% of the variance in log₁₀ C over three orders of magnitude in pigment concentration. This is combined with solutions to the radiative transfer equation to simulate the ocean's apparent spectral radiance at satellite altitudes as a function of C and the optical properties of the aerosol, the optical depth of which is assumed to be proportioned to λ^-n. A specific atmospheric correction algorithm, based on the assumption that the ocean is totally absorbing at 670 nm, is then applied to the simulated spectral radiance, from which the pigment concentration is derived. Comparison between the true and derived values of C show that: (1) n is considerably more important than the actual aerosol optical thickness; (2) for C 0299-1 0.2 Μg l^-1 acceptable concentrations can be determined as long as n is not overestimated; (3) as C increases, the accuracy with which n must be estimated, for a given relative accuracy in C, also increases; and (4) for C greater than about 0.5 Μg 1^-1, the radiance at 440 nm becomes essentially useless in determining C. The computations also suggest that if separate pigment algorithms are used for C ≲ 1Μgl^-1 and C ≳ 1 Μgl^-1, accuracies considerably better than ±± in log C can be obtained for C ≲ 1 Μg l^-1 with only a coarse estimate of n, while for C ≳ 10 Μgl^-1, this accuracy can be achieved only with very good estimates of n.